Guide

Requirements

• Python 3.11+
• An Anthropic API key
• A vault directory — a folder of Markdown files (Obsidian works directly)

Optional — required only if you configure a local model via Ollama:

• Ollama installed and running
• pip install openai (outheis uses the OpenAI-compatible Ollama API)

Optional for messaging via Signal:

• A registered Signal account for the bot phone number
• signal-cli installed and configured

Quickstart

The fastest path to a working setup is an Anthropic API key. No additional dependencies — outheis works out of the box with Claude as the only model provider.

Local models via Ollama are optional and require extra setup (see below).

Installation

Modern pip enforces virtual environments. Create one first:

git clone https://github.com/outheis-labs/outheis-minimal.git
cd outheis-minimal
python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -e .

The outheis command is only available while the venv is active. Add source /path/to/outheis-minimal/.venv/bin/activate to your shell profile if you want it always available.

For voice transcription via Signal, install the optional dependency:

pip install -e ".[signal]"

Setup

outheis init

This creates ~/.outheis/human/config.json with defaults. Edit it:

$EDITOR ~/.outheis/human/config.json

Minimal required configuration:

{
  "human": {
    "name": "Your Name",
    "language": "en",
    "timezone": "Europe/Berlin",
    "vault": ["~/Documents/Vault"]
  },
  "llm": {
    "providers": {
      "anthropic": {
        "api_key": "sk-ant-..."
      }
    },
    "models": {
      "fast":    {"provider": "anthropic", "name": "claude-haiku-4-5"},
      "capable": {"provider": "anthropic", "name": "claude-sonnet-4-20250514"}
    }
  }
}

The vault array accepts multiple paths. outheis monitors all of them.

Starting the Dispatcher

outheis start        # Background daemon
outheis start -f     # Foreground (useful for first run / debugging)
outheis status       # Check PID, uptime, agent status
outheis stop         # Stop daemon

Web UI

Once the dispatcher is running, the Web UI is available at:

http://127.0.0.1:8080

It provides:

• Live message feed (conversations with agents)
• Memory, Rules, and Skills viewer and editor
• Scheduler status and manual task triggers
• Vault file browser

The Web UI port and host are configurable in config.json under "webui": {"host": "127.0.0.1", "port": 8080}.

Remote access via SSH

The Web UI binds to localhost only. To access it from another machine, use SSH port forwarding:

ssh -L 8080:localhost:8080 user@your-server

Then open http://localhost:8080 in your local browser. The tunnel stays open as long as the SSH connection is active.

Vault Setup

Your vault is your primary knowledge store. outheis reads from it, writes to it (via cato and the Web UI), and learns from it over time. Structure is flexible — any Markdown files work. The recommended Agenda layout:

vault/
└── Agenda/
    ├── Agenda.md     # Today's schedule — written by cato
    ├── Inbox.md      # Your quick capture — processed hourly
    └── Exchange.md   # Async back-and-forth with cato

Create the Agenda/ directory and empty files. cato generates Agenda.md on first run.

CLI Commands

Daemon Control

outheis start       # Start dispatcher (background)
outheis start -f    # Start in foreground
outheis start -fv   # Foreground + verbose (shows tool calls)
outheis stop        # Stop dispatcher
outheis status      # Show status, PID, uptime

Messaging

outheis send "Hello"              # Single message
outheis send "@zeno find notes"   # Direct to Data agent
outheis chat                      # Interactive mode (with history)

Memory

outheis memory              # Show all memories
outheis memory --type user  # Show only user facts

Rules

outheis rules         # Show all rules (system + user)
outheis rules relay   # Show relay agent rules

Talking to outheis

Just talk naturally. Relay decides when to use tools:

You say	What happens
"hi"	Direct response
"what's on today?"	Uses check_agenda tool → Agenda agent
"where do I live?"	Uses search_vault tool → Data agent
"! I am 54"	Saves to Memory (explicit marker)
"@ Call dentist Friday"	Writes directly to Agenda

Prefixes

Prefix	Meaning	Example
!	Save to Memory	! I prefer short answers
@	Write to Agenda	@ Team meeting Thursday 10am

The @ prefix (at-sign followed by a space) bypasses relay entirely — the item is passed directly to the agenda agent without clarifying questions.

Explicit Agent Mentions

Use @name (no space) for direct delegation:

Mention	Agent	Use for
@zeno	Data	Search vault explicitly
@cato	Agenda	Schedule queries
@hiro	Action	External actions (future)

Vault

Your vault is a directory of Markdown files:

---
title: Project Alpha
tags: [active, client-work]
created: 2025-01-15
---
# Project Alpha

Status update...

Recommended Structure

vault/
├── Agenda/
│   ├── Agenda.md     # Today's schedule
│   ├── Inbox.md      # Unprocessed items
│   └── Exchange.md   # External sync
├── projects/
├── notes/
└── references/

Configuration

~/.outheis/human/config.json:

{
  "user": {
    "name": "string",
    "language": "en|de|...",
    "timezone": "Region/City",
    "vault": ["~/path/to/vault"]
  },
  "llm": {
    "provider": "anthropic"
  }
}

Troubleshooting

"Dispatcher not running"

outheis status   # Check if running
outheis start    # Start it

Stale PID file

rm ~/.outheis/.dispatcher.pid
outheis start

No API key

export ANTHROPIC_API_KEY="sk-ant-..."
# Or add to ~/.bashrc / ~/.zshrc

"openai package not installed" (Ollama models)

If you configure a model with "provider": "ollama", outheis uses the OpenAI-compatible Ollama API and requires the openai Python package:

pip install openai

Also make sure Ollama itself is running (ollama serve) and the model is pulled (ollama pull <model>).

macOS: Daemon won't start in background

Use foreground mode:

outheis start -f &